AI lung scan score links to fibrosis progression and mortality
A baseline deep learning score above 0.5 was associated with a 13.41% relative annual fibrosis increase and higher mortality in two human cohorts.
American Journal of Respiratory and Critical Care Medicine
In human participants without clinically diagnosed interstitial lung disease, automated CT scan analysis identified individuals at elevated risk of lung fibrosis progression and death. Researchers analyzed baseline and follow-up CT scans from two observational cohorts, COPDGene and AGES-Reykjavík. They applied data-driven textural analysis (DTA) to quantify fibrosis and a deep learning classifier named MIL-UIP to detect patterns resembling usual interstitial pneumonia.
A baseline MIL-UIP score above 0.5 was associated with a 13.41% relative annual increase in DTA-measured fibrosis (95% CI: 8.10% to 18.99%). The elevated score was also associated with increased mortality in both COPDGene (hazard ratio 1.61; 95% CI: 1.08 to 2.38) and AGES-Reykjavík (hazard ratio 1.62; 95% CI: 1.01 to 2.61). Baseline scores strongly tracked with visual assessments of interstitial lung abnormalities.
Why it matters
Interstitial lung abnormalities become increasingly prevalent with age and can precede fatal pulmonary fibrosis. Automated imaging biomarkers may help identify which older adults face the highest risk of declining respiratory health before clinical disease develops.
Caveats
The study relied on observational data from two specific cohorts, which limits causal conclusions. Prospective validation is still required before using these automated classifiers for clinical trial enrichment or routine patient care.
The paper
Show 8 more authors
Gary Matthew Hunninghake, Jonathan Rose, Rachel K. Putman, Gunnar Guðmundsson, Vilmundur Gudnason, Joshua J. Solomon, Soonho Yoon, David A. Lynch,National Jewish Health
American Journal of Respiratory and Critical Care Medicine · 29 Sep 2026